CN106291381A - A kind of Combined estimator electrokinetic cell system state-of-charge and the method for health status - Google Patents

Abstract

The application relates to a kind of method of Combined estimator electrokinetic cell system state-of-charge and health status, utilize Multiple Time Scales filtering algorithm, use macroscopic time yardstick to obtain electrokinetic cell system estimates of parameters, use microcosmic time scale estimating system state, assess health status and the state-of-charge of described electrokinetic cell.Form electrokinetic cell parameter based on Multiple Time Scales and the combined estimation method of state, it is achieved electrokinetic cell active volume and the state-of-charge accurate Combined estimator in uncertain applied environment.Not only make estimated result more reliable and more stable, reduce the calculating cost of system simultaneously.

Description

A kind of Combined estimator electrokinetic cell system state-of-charge and the method for health status

Technical field: the present invention relates to technical field of power battery management, especially vehicle mounted dynamic battery systematic parameter is distinguished Know the method for estimation with state estimation.

Background technology:

At present, SOC algorithm based on model needs the SOC using Kalman filtering serial algorithm to estimate battery mostly, Although theory shows that Kalman filtering is the linear state estimation algorithm of a kind of optimum, but to be only applicable to model accurate, defeated for it Entering situation known to noise statistics, this obviously can not meet the demand of reality.

Based on this, H of the present invention_∞Filtering is a kind of algorithm aiming at robustness and design.This algorithm mainly side Formula sees " optimal State Estimation Kalman, H_∞And nonlinear filtering ", National Defense Industry Press.It is different from Kalman's filter Ripple, even if the input statistical property in the existence error of model, noise is unknown, the most in the worst cases, this algorithm is still State estimation can be accurately finished.

The On-line Estimation method of existing electrokinetic cell SOC, the most true due to maximum available (hereinafter referred to as capacity) The qualitative estimated result reliability making electrokinetic cell SOC is low.And it is existing by metastable SOC-OCV curve (electricity of i.e. opening a way Buckle line) as the fair curve of SOC algorithm for estimating, battery, when using temperature, degree of aging etc. different, this curve equally can Occur the most significantly to change.

Based on this, the present invention in view of different degree of agings, at a temperature of OCV-SOC curve all can produce necessarily with capacity Change, thus set up the three-dimensional response surface of capacity, state-of-charge and open-circuit voltage, i.e. capacity-SOC-OCV three-dimensional response Face, achieves the Combined estimator of the different SOC used under environment (degree of aging, temperature) and capacity based on this.

Present invention is alternatively directed to the quick time-varying characteristics of electrokinetic cell system quantity of state and the slow time-varying characteristics of parameter amount, adopt With the SOC of microcosmic time scale estimated driving force battery, use the model parameter of macroscopic time size estimation electrokinetic cell with available Capacity, forms electrokinetic cell SOC based on Multiple Time Scales and the combined estimation method of capacity, it is achieved electrokinetic cell SOC and appearance Measure the accurate Combined estimator in uncertain applied environment.Not only make estimated result more reliable and more stable, reduce simultaneously The calculating cost of system.

Summary of the invention:

A kind of method that the present invention relates to Combined estimator electrokinetic cell system state-of-charge and health status, described method bag Include:

First, setting up the capacity-SOC-OCV three-dimensional response surface, described OCV is the open-circuit voltage of described system；

Secondly, online data obtains, Real-time Collection electrokinetic cell monomer and the voltage and current of power battery pack；

Then, Multiple Time Scales filtering algorithm, it is thus achieved that under the current macroscopic time scale of described electrokinetic cell system be System capacity advance estimate modification value and the system SOC advance estimate modification value under current microcosmic time scale；

Under each SOC estimates microcosmic sampled point, update described system SOC advance estimate modification value, estimate every L described SOC Microcosmic sampled point is capacity estimation macroscopic view sampled point, updates described power system capacity and estimate and repair under this capacity estimation macroscopic view sampled point On the occasion of, the described power system capacity advance estimate modification value after every time updating is as the renewal of L time after current capacities estimation macroscopic view sampled point Parameter used by described system SOC advance estimate modification value；

Described L is two or more；

Finally, online SOC Yu SOH extracts, and utilizes presently described system SOC that described Multiple Time Scales filtering algorithm obtains Advance estimate modification value and power system capacity advance estimate modification value, estimated driving force battery system state-of-charge and health status.

Preferably, described Multiple Time Scales filtering algorithm includes:

Step is 1.: carries out current SOC and estimates that system SOC under microcosmic sampled point k is estimated, obtains the discreet value of system SOC；

Step is 2.: based on step 1. in the system SOC discreet value of acquisition and last capacity estimation macroscopic view sampled point be System capacity advance estimate modification value, utilizes the described capacity-SOC-OCV three-dimensional response surface, updates current system open-circuit voltage and obtains first Open-circuit voltage；

Then, based on above-mentioned first open-circuit voltage, carry out the correction of described system SOC discreet value, obtain described system SOC Advance estimate modification value；

Step is 3.: k+1 estimates microcosmic sampled point as new SOC, it is judged that whether k-1 can be divided exactly by L, if it can, then enter Row step is 4.；Otherwise return step 1.；

Step is 4.: the power system capacity carried out under capacity estimation macroscopic view sampled point l+1 is estimated, and obtains power system capacity discreet value；

Step is 5.: based on step 4. in the power system capacity discreet value of acquisition and the last step 2. in the system that obtains SOC advance estimate modification value, utilizes the described capacity-SOC-OCV three-dimensional response surface, again updates current system open-circuit voltage and obtains second Open-circuit voltage；

Based on described second open-circuit voltage, carry out repairing of the power system capacity discreet value under capacity estimation macroscopic view sampled point l+1 Just, power system capacity advance estimate modification value is obtained；And return step 1..

Preferably, step 1. in utilize SOC to estimate the System current under microcosmic sampled point k-1 and system SOC are estimated and are repaiied On the occasion of, and the power system capacity advance estimate modification value of the last sampled point l carries out system SOC and estimates.

Preferably, step 2. in carry out the correction of described system SOC discreet value, based on step 1. in the system of acquisition SOC discreet value, described first open-circuit voltage, SOC estimate the current value under microcosmic sampled point k and magnitude of voltage.

Preferably, step 5. in carry out the correction of described power system capacity discreet value, based on the second open-circuit voltage, capacity estimation Macroscopic view sampled point l+1 or the last SOC estimates the current value under microcosmic sampled point and magnitude of voltage.

Preferably, power system capacity advance estimate modification value is utilized to assess described SOH.

Preferably, described SOC estimates that sample frequency and the described electric current of microcosmic sampled point, voltage sample frequency are identical.

Preferably, setting up the method for described capacity-SOC-OCV three-dimensional response surface is: to SOC and OCV under different capabilities Relation is fitted, and obtains the built-up pattern coefficient under each described different capabilities, uses quadratic function to each built-up pattern coefficient It is fitted with the relation of capacity, completes the foundation of capacity-SOC-OCV three-dimensional response surface.

Preferably, by battery management system Real-time Collection electrokinetic cell monomer and the voltage and current of power battery pack.

Combined estimation method proposed by the invention has the advantage that compared with traditional method

(1) expected value using SOC to estimate is more complete with the limits of error (as a example by 95% confidence interval, but being not limited to 95%) Face, the state-of-charge evaluating electrokinetic cell exactly may distribution situation；

(2) in the case of SOC with capacity initial value the most inaccurate (20% error), all energy rapid convergences, to true value, i.e. achieve During battery heap(ed) capacity the unknown, the accurate estimation of SOC, solve before tradition SOC algorithm for estimating is known as with maximum available Carry and cannot Successful utilization to the difficult problem on real vehicle；

(3), for comparing single time scale algorithm, not only estimated accuracy is greatly improved, and reduces significantly Algorithm amount of calculation and the time of calculating.

Accompanying drawing illustrates:

Electrokinetic cell system SOC and the SOH combined estimation method flow chart of Fig. 1 Multiple Time Scales；

Fig. 2 electrokinetic cell Thevenin equivalent-circuit model；

Fig. 3 capacity-SOC-OCV three-dimensional response surface；

The double H of the mono-time scale of Fig. 4_∞Filtering algorithm terminal voltage, SOC and capacity estimation result.Wherein: (a), terminal voltage are predicted Value contrasts with measured value；(b), terminal voltage forecast error；C (), SOC estimation contrast with reference value；(d), SOC estimation difference； E (), capacity estimation value contrast with reference value；(f), capacity estimation error.

Fig. 5 Multiple Time Scales H_∞Filtering algorithm terminal voltage, SOC and capacity estimation result.Wherein: (a), terminal voltage predictive value Contrast with measured value；(b), terminal voltage forecast error；C (), SOC estimation contrast with reference value；(d), SOC estimation difference； E (), capacity estimation value contrast with reference value；(f), capacity estimation error.

Detailed description of the invention:

As it is known in the art, the electrokinetic cell system of the present invention includes electrokinetic cell monomer, electrokinetic cell bag or becomes Electrokinetic cell after group.

The present invention uses OCV to represent open-circuit voltage, and SOC represents battery charge state, and SOH represents cell health state.

A kind of electrokinetic cell system state-of-charge (SOC) based on Multiple Time Scales of the present invention and health status (SOH) combined estimation method is as shown in Figure 1.

Electrokinetic cell system state-of-charge (SOC) based on Multiple Time Scales of the present invention and the connection of health status (SOH) Close method of estimation and be applied equally to nonlinear system.

The state of present system refers to the system index changed constantly, including battery SOC, polarizing voltage, such as SOC once Charge and discharge process i.e. completes the change of a complete cycle completely.And the parameter of system refers to slower relative to state change System index, such as battery capacity and battery model parameter, it is once having almost no change in charge and discharge process completely.This The explanation of bright following system no special all refers to electrokinetic cell system, system mode correspondence battery system state, preferably corresponds to electricity The SOC of cell system.Systematic parameter correspondence battery system parameter, preferably corresponds to battery system SOH or maximum available.

This combined estimation method includes: the foundation of capacity-SOC-OCV three-dimensional response surface, online data obtain, chi of many time Degree H_∞Filtering algorithm and online SOC Yu SOH extract four aspects.Separately below aforementioned four aspect is described the most in detail:

Algorithm preparation: the foundation of capacity-SOC-OCV three-dimensional response surface

Metastable SOC-OCV curve, i.e. open circuit voltage curve, often as the fair curve of SOC algorithm for estimating, but When battery uses the change such as temperature, degree of aging, this curve can occur the most significantly to change equally.The present invention is by temperature, old The impact of this curve is directly reflected on the difference of battery capacity by the factors such as change degree, utilizes capacity, SOC Yu OCV three Relation as the correction curved surface of capacity Yu SOC Combined estimator algorithm.Detailed process is as follows:

Open voltage test is carried out, to obtain different electricity under different battery capacities (i.e. when temperature, degree of aging change) SOC Yu OCV corresponding relation under tankage, use built-up pattern (as shown in formula (1)) respectively to different capabilities under SOC and OCV relation is fitted, thus obtains the α under each different capabilities₀,α₁,…,α₆Parameter value, finally use quadratic function (as Shown in formula (2)) to parameter alpha₀,α₁,…,α₆It is fitted with the relation of capacity, so far completes the capacity-SOC-OCV three-dimensional response surface Foundation.

U_oc(C_a, z)=α₀+α₁z+α₂z²+α₃z³+α₄/z+α₅ln(z)+α₆ln(1-z) (1)

${\left[\begin{array}{ccccccc}{\mathrm{\α}}_0& {\mathrm{\α}}_1& {\mathrm{\α}}_2& {\mathrm{\α}}_3& {\mathrm{\α}}_4& {\mathrm{\α}}_5& {\mathrm{\α}}_6\end{array}\right]}^T=\mathrm{\Λ}\×{\left[\begin{array}{ccc}{C}_a^2& C_a& 1\end{array}\right]}^T---\left(2\right)$

C_aFor battery capacity；

Z is battery SOC；

U_oc(C_a, z) representing open-circuit voltage (OCV), it is the function of battery capacity and SOC；

α₀,α₁,…,α₆Coefficient for built-up pattern；

The transposition of subscript T representing matrix；

Λ is 7 × 3 constant matricess.

1, online data obtains

When electric automobile runs, the battery management system (BMS) in electrokinetic cell system can pass through data acquisition unit The information such as Real-time Collection electrokinetic cell monomer and the voltage of power battery pack, electric current, and it is stored in corresponding memorizer, for following Multiple Time Scales H_∞Filtering algorithm provides reliably real time information to input, and the input of described information includes t_kTime etching system measurement Value y_k=U_t,k, t_kTime etching system input information u_k=i_L,k。

i_L,kFor controlling electric current；

U_t,kFor terminal voltage.

2, Multiple Time Scales H_∞Filtering algorithm

The present invention uses Multiple Time Scales H_∞Filtering algorithm realizes electrokinetic cell parameter and accurately estimates with state joint.Should Algorithm is applicable to following nonlinear discrete systems:

$\left\{\begin{array}{c}{x}_k=f(x_{k-1},{\mathrm{\θ}}_l,u_{k-1})+w_{k-1}\\ {\mathrm{\θ}}_l={\mathrm{\θ}}_{l-1}+{\mathrm{\ρ}}_{l-1}\\ y_k=g(x_k,{\mathrm{\θ}}_l,u_k)+v_k\end{array}\right.---\left(3\right)$

X represents the state of system；

θ represents the parameter of system；

Subscript k represents t_kMoment systematic sampling time point (extraneous input), also represent the time chi of state estimation simultaneously Degree, i.e. all carries out a state estimation under each sampling time puts；

Subscript l represents the time scale of parameter estimation, and it is numerically equal to k business's (L is spatial scaling limit value) divided by L, I.e. carry out primary parameter identification every L sampling time point, and parameter identification result is used to estimate t every time_l×LAfter moment The state value inscribed when L；

Microcosmic time scale, the time scale of the most above-mentioned state estimation；

The time scale that macroscopic time yardstick, i.e. above-mentioned parameter are estimated；

f(x_k-1,θ_l,u_k-1) represent model function of state；

g(x_k,θ_l,u_k) represent model observation function；

y_kFor t_kTime etching system measured value, y_k=U_t,k；

u_kFor t_kTime etching system input information, u_k=i_L,k；

w_k-1And ρ_l-1It is respectively system mode noise and parametric noise, v_kFor measuring noise, at H_∞Among filtering algorithm, institute State system mode noise, parametric noise and measurement noise and be designed to random and unknown, breach tradition filtering algorithm state Noise, parametric noise and measurement noise be white noise this it is assumed that thus be combined tightr with actual production.

For this nonlinear discrete systems of battery system, above-mentioned parameter is defined as follows:

The present invention illustrates as a example by using Thevenin electrokinetic cell equivalent-circuit model that this electrokinetic cell SOC Yu SOH joins Close method of estimation.Fig. 2 is Thevenin electrokinetic cell equivalent-circuit model, and this model is by voltage source, ohmic internal resistance and RC net Network three part forms.Corresponding mathematics model is set up, as shown in formula (4) according to each part characteristic and electricity philosophy.

$\left\{\begin{array}{c}{\stackrel{\·}{U}}_p=-\frac{1}{C_p{R}_p}{U}_p+\frac{1}{C_p}{i}_L\\ U_t=U_{oc}-U_p-i_L{R}_0\end{array}\right.---\left(4\right)$

U_pFor polarizing voltage,For its derivative；

C_pFor polarization capacity；

R_pFor polarization resistance；

i_LFor controlling electric current；

U_tFor terminal voltage；

U_ocFor open-circuit voltage；

R₀For ohmic internal resistance.

The accounting equation of electrokinetic cell SOC is:

$z=z_0-\frac{\∫i_{L}dt}{C_a}---\left(5\right)$

z₀Represent the initial value of SOC；

C_aFor electrokinetic cell maximum available (hereinafter referred to as capacity), i.e. refer under the conditions of certain use, battery The maximum electricity can released after fully charged, battery maximum available is the important ginseng characterizing cell health state (SOH) simultaneously Number, under the conditions of the most identical use, battery maximum available is the least, and battery decay is the most obvious, and cell health state (SOH) is more Difference.

In view of the sampling of voltage x current is discrete, i.e. the input of SOC algorithm for estimating is also discrete, thus be necessary by Formula (4) and (5) linear discrete, be rewritten into simultaneously and comprise hidden state x as shown in formula (6)_kWith implicit parameter θ_lMany times Yardstick electrokinetic cell system, it may be assumed that

$\begin{array}{c}\left\{\begin{array}{c}{x}_k=f(x_k,{\mathrm{\θ}}_l,u_k)+w_k=\left[\begin{array}{cc}\exp \left(\frac{-\mathrm{\Δ}t}{C_{p,l}{R}_{p,l}}\right)& 0\\ 0& 1\end{array}\right]x_k+\left[\begin{array}{c}\[1-\exp \left(\frac{-\mathrm{\Δ}t}{C_{p,l}{R}_{p,l}}\right)\]i_{L,k}{R}_{p,l}\\ \mathrm{\η}\left(i_{L,k}\right)i_{L,k}\mathrm{\Δ}t/C_{a,l}\end{array}\right]+w_k\\ {\mathrm{\θ}}_{l+1}={\mathrm{\θ}}_l+{\mathrm{\ρ}}_l\\ y_k=g(x_k,{\mathrm{\θ}}_l,u_k)=U_{oc,k}-U_{p,k}-R_{0,l}{i}_{L,k}+v_k\end{array}\right.\\ \left\{\begin{array}{c}{x}_k={\left[\begin{array}{cc}{U}_{p,k}& z_k\end{array}\right]}^T\\ {\mathrm{\θ}}_l={\left[\begin{array}{cccc}{R}_{0,l}& R_{p,l}& C_{p,l}& C_{a,l}\end{array}\right]}^T\\ y_k=U_{t,k}\\ u_k=i_{L,k}\end{array}\right.\end{array}---\left(6\right)$

The unit interval of Δ t express time yardstick k；

η(i_L,k-1) represent efficiency for charge-discharge.

Based on above formula (1), (2) and (6), obtain double H_∞In filtering algorithm, function of state and observation function are about state or ginseng The Jacobian matrix of number:

$A_{k-1}=\frac{df(x,{\mathrm{\θ}}_l,u_{k-1})}{dx}=\left[\begin{array}{cc}\exp \left(\frac{-\mathrm{\Δ}t}{C_{p,l}{R}_{p,l}}\right)& 0\\ 0& 1\end{array}\right]---\left(7\right)$

$C_k^x=\frac{dg(x,{\mathrm{\θ}}_l,u_k)}{dx}={\[-1\frac{\∂U_{oc}(C_{a,l},z)}{\∂z}\]}^T---\left(8\right)$

$C_l^{\mathrm{\θ}}=\frac{dg(x_k,\mathrm{\θ},u_k)}{d\mathrm{\θ}}={\left[\begin{array}{cccc}-i_{L,k}& 0& 0& \frac{\∂U_{oc}(C_a,z_k)}{\∂C_a}\end{array}\right]}^T+C_k^x\frac{\∂x_k}{\∂\mathrm{\θ}}---\left(9\right)$

A_k-1For function of state about the Jacobian matrix of state；

For observation function about the Jacobian matrix of state；

For observation function about the Jacobian matrix of parameter.

Here, function derivation result each in above formula (7-9) is arranged, it may be assumed that

$\frac{\∂U_{oc}}{\∂z}={\mathrm{\α}}_1+2{\mathrm{\α}}_{2}z+3{\mathrm{\α}}_3{z}^2-{\mathrm{\α}}_4/z^2+{\mathrm{\α}}_5/z-{\mathrm{\α}}_6/(1-z)---\left(10\right)$

$\frac{\∂U_{oc}}{\∂C_a}=\left[\begin{array}{ccccccc}1& z& z^2& z^3& 1/z& ln\left(z\right)& ln(1-z)\end{array}\right]\×\mathrm{\Λ}\×\left[\begin{array}{ccc}2C_a& 1& 0\end{array}\right]---\left(11\right)$

$\frac{{\mathrm{dx}}_k}{d\mathrm{\θ}}=\frac{\∂f(x_{k-1},u_{k-1},\mathrm{\θ})}{\∂\mathrm{\θ}}+A_{k-1}\frac{{\mathrm{dx}}_{k-1}}{d\mathrm{\θ}}---\left(12\right)$

Wherein, initial value dx₀/ d θ, in the case of failing to obtain believable empirical value, is usually initialized to zero.

$\frac{\∂f(x_{k-1},u_{k-1},\mathrm{\θ})}{\∂\mathrm{\θ}}=\left[\begin{array}{cccc}0& b_1& b_2& 0\\ 0& 0& 0& b_4\end{array}\right]---\left(13\right)$

$\left\{\begin{array}{c}{b}_1=\frac{U_{p,k-1}\mathrm{\Δ}t}{R_{p,l}^2{C}_{p,l}}\exp \left(\frac{-\mathrm{\Δ}t}{C_{p,l}{R}_{p,l}}\right)-\frac{i_{L,k-1}\mathrm{\Δ}t}{C_{p,l}{R}_{p,l}}\exp \left(\frac{-\mathrm{\Δ}t}{C_{p,l}{R}_{p,l}}\right)-i_{L,k}\[\exp \left(\frac{-\mathrm{\Δ}t}{C_{p,l}{R}_{p,l}}\right)-1\]\\ b_2=\frac{U_{p,k-1}\mathrm{\Δ}t}{C_{p,l}^2{R}_{p,l}}\exp \left(\frac{-\mathrm{\Δ}t}{C_{p,l}{R}_{p,l}}\right)-\frac{i_{L,k-1}\mathrm{\Δ}t}{C_{p,l}^2}\exp \left(\frac{-\mathrm{\Δ}t}{C_{p,l}{R}_{p,l}}\right)\\ b_4=\frac{\mathrm{\η}\left(i_{L,k}\right)i_{L,k}\mathrm{\Δ}t}{C_{a,l}^2}\end{array}\right.---\left(14\right)$

So far, the definition of each relevant parameter in the non-linear off-line system of electrokinetic cell has been completed, as shown in formula (6)-(9). Below this algorithm detailed process is described:

The initialization of algorithm: be respectively provided with macroparameter observer H_∞F_θWith micro state observer H_∞F_xInitial parameter Value.Including:

${\widehat{\mathrm{\θ}}}_0^{+},Q_0^{\mathrm{\θ}},R_0^{\mathrm{\θ}},{P_0^{\mathrm{\θ}}}^{,+},S_{\mathrm{\θ}},{\mathrm{\λ}}_{\mathrm{\θ}}---\left(15\right)$

${\hat{x}}_0^{+},Q_0^x,R_0^x,P_0^{x,+},S_x,{\mathrm{\λ}}_x,{\mathrm{\Σ}}_0^{x,+}$

For parameter estimator device H_∞F_θMiddle systematic parameterInitial value；

Represent parameter θ_kEstimated value or expected value, i.e.

For parameter estimator device H_∞F_θMiddle matrixInitial value；

For designer based on system noise ρ_lDesigned symmetric positive definite matrix, if being aware of ρ in advance such as us_k? During three element very big (such as several orders of magnitude bigger than other element), thenShould be more thanIn other element；

For parameter estimator device H_∞F_θMiddle matrixInitial value；

For designer based on measuring noise v_kDesigned symmetric positive definite matrix, if being aware of v in advance such as us_k? When two elements are the biggest, thenShould be more thanIn other element；

S_θFor parameter estimator device H_∞F_θMiddle designer degree of attentiveness based on component each in parameter θ and the symmetric positive definite that sets Battle array, as when we are to state vector θ_kThe 3rd element interested time, then S can be designed_θ(3,3) make it the biggest In S_θIn other element；

For parameter estimator device H_∞F_θMiddle designer is based on the symmetric positive definite matrix designed by system estimation error, as when me To parameter vector initial value θ₀The 3rd element when knowing nothing, then can designIt is far longer than In other element；

λ_θFor parameter estimator device H_∞F_θSelected performance bounds, selected performance bounds value the biggest explanation algorithm robustness is more By force, the interference (such as noise etc.) in the external world can be better adapted to, and when performance bounds is set to 0 (minima), algorithm is degenerated For Kalman filtering algorithm, but big performance bounds value tends to rely on matrixWithAbundant appropriate design so that The debugging difficulty of algorithm is bigger；

For state observer H_∞F_xIn system modeInitial value；

It is state x_kEstimated value or expected value, i.e.

For state observer H_∞F_xIn system state estimation error co-variance matrixInitial value；

For state estimation covariance matrix, different from the past only from expected valueEvaluating estimated result, this patent is from expectation ValueAnd varianceTwo aspects more fully evaluate system estimation result, and introduce estimation difference limit (formula (12)) concept, make Obtain estimated result more directly perceived；

For state observer H_∞F_xMiddle matrixInitial value；

For designer based on system noise w_kDesigned symmetric positive definite matrix, if being aware of w in advance such as us_k? When two elements are the biggest, thenShould be more thanIn other element；

For state observer H_∞F_xMiddle matrixInitial value；

For designer based on measuring noise v_kDesigned symmetric positive definite matrix, if being aware of v in advance such as us_k? When three elements are the biggest, thenShould be more thanIn other element；

S_xFor parameter estimator device H_∞F_xMiddle designer degree of attentiveness based on component each in state x and the symmetric positive definite that sets Battle array, as when we are to state vector x_kThe 3rd element interested time, then S can be designed_x(3,3) make it far away More than S_xIn other element；

For parameter estimator device H_∞F_xMiddle designer is based on the symmetric positive definite matrix designed by system estimation error, as when me To parameter vector initial value x₀The 3rd element when knowing nothing, then can designIt is far longer thanIn Other element；

λ_xFor state observer H_∞F_xSelected performance bounds, selected performance bounds value the biggest explanation algorithm robustness is more By force, the interference (such as noise etc.) in the external world can be better adapted to, and when performance bounds is set to 0 (minima), algorithm is degenerated For Kalman filtering algorithm, but big performance bounds value tends to rely on matrixWithAbundant appropriate design, thus now The debugging difficulty of algorithm is bigger.

When sampling time k ∈ 1,2 ..., ∞ time, based on the information such as electric current, voltage continually enter, calculate:

Step is 1.: state observer H based on microcosmic time scale_∞F_xTime update (prior estimate)

Utilize the current value u of k-1 sampled point_k-1=i_L,k-1, the correction value of system state estimation value, and the last time adopt The correction value of the systematic parameter estimated value of sampling point lCarry out system mode to estimate.

Meanwhile, the system mode design matrix utilizing k-1 sampled point is estimatedThe time of completion system Design of State matrix UpdateDescribed system mode design matrix updatesStep 2. in be used to update state gain matrixSuch as formula (20) shown in.

System mode is estimated:

System mode design matrix is estimated:

System mode error covariance is estimated: For t_kMoment state priori estimates,I.e. use t_kThe measured value (not including k) before moment comes Estimate x_k；

Represent the function of state of model；

For t_k-1Moment state posterior estimate,I.e. use t_k-1Moment and t_k-1Moment X estimated by measured value in the past_k-1, circulate (i.e. t based on the last time here_k-1Moment) output directly obtain；

For t_l×LMoment parameter posterior estimate,I.e. use t_l×LMoment and t_l×LMoment with θ estimated by front measured value_l, output based on previous loops here directly obtains；

u_kFor t_kTime etching system input information, it is known quantity；

For t_kMoment Design of State matrixPrior estimate result, withCorresponding；

For withFor initial value and by formula (17), (22) H that constantly recursion obtains_∞Infinite filter state design matrix；For t_k-1The Design of State matrix in momentPosterior estimator result, withCorresponding, circulate (i.e. based on the last time here t_k-1Moment) output directly obtain；

For designer based on system noise w_kDesigned symmetric positive definite matrix is at t_k-1The value in moment；

For t_kMoment state estimation error co-variance matrixPrior estimate result, withCorresponding；

For t_k-1The state estimation error co-variance matrix in momentPosterior estimator result, withCorresponding, this In based on the last time circulate (i.e. t_k-1Moment) output directly obtain.

Step is 2.: state observer H based on microcosmic time scale_∞F_xMeasurement updaue (Posterior estimator)

Based on step 1. in the system mode of acquisition estimateEstimate with the systematic parameter of last parameter estimation sampled point Value correction value(i.e. parameter posterior estimate), utilizes the capacity-SOC-OCV three-dimensional response surface, it is thus achieved that the first corresponding open circuit electricity Pressure U_oc。

Then, based on above-mentioned U_ocWith the electric current u under current sampling point_k=i_L,kWith magnitude of voltage y_k=U_L,k, carry out current shape State estimates that the state measurement under microcosmic sampled point k updates, i.e. system state estimation value correction shown in formula (21).

Wherein as shown in formula (6), U_ocWith directly determineSize, utilize U_ocComplete formula (19) institute Show that system state estimation newly ceases renewal.

System state estimation newly ceases renewal:

System mode gain matrix:

$K_k^x=P_k^{x,-}{(I-{\mathrm{\λ}}_x{S}_k^x{P}_k^{x,-}+{\left(C_k^x\right)}^T{\left(R_k^x\right)}^{-1}{C}_k^x{P}_k^{x,-})}^{-1}{\left(C_k^x\right)}^T{\left(R_k^x\right)}^{-1}---\left(20\right)$

Estimate in system modeOn the basis of, utilize system state estimation newly to cease renewalWith system mode gain matrixRightIt is modified, obtains system state estimation value correction

System state estimation value correction:

Wherein, the system mode design matrix discreet value that 1. step obtains is utilizedCompletion system Design of State matrix Measurement updaueDescribed system mode design matrix updatesIt is used to estimate the system mode design matrix of subsequent timeSee formula (17).

System mode design matrix updates:

Due to end-state estimated result present withFor expected value,For the approximate normal distribution of variance, hence with State estimation error covariance evaluates degree of accuracy and the stability of state estimation.

System state estimation error covariance updates:

The system state estimation limits of error (95% confidence interval): Newly cease for state estimation, i.e. the predictor error of measured value；

y_kFor t_kTime etching system measured value；

Represent the observation function of model；

For state gain matrix；

For designer based on measuring noise v_kDesigned symmetric positive definite matrix is at t_kThe value in moment；

I is unit matrix；

α is constant vector, is used for extracting matrix(on this diagonal of a matrix, element is much larger than its other element of being expert at) is right Element on linea angulata, and α=[1 1 1]^T。

Step is 3.: judges whether k-1 can be divided exactly by L, if it can, then make l=l+1, and continues next step；Otherwise, then return Return the calculating preparing next sampling instant.

Step is 4.: state observer H based on macroscopic time yardstick_∞F_θTime update (prior estimate)

Systematic parameter is estimated:

System Parameter Design matrix is estimated: For t_l×LMoment parameter priori estimates,I.e. use t_l×L(t was not included before moment_l×L) measurement Value estimates θ_l；

For t_l×LMoment Design of State matrix P_l ^θPrior estimate result, withCorresponding；

P_l ^θFor withFor initial value and by formula (26), (30) H that constantly recursion obtains_∞Infinite Filter Parameter Design square Battle array；For t_(l-1)×LThe Design of State matrix in momentPosterior estimator result, withCorresponding, follow here by the last time The result of calculation of ring directly obtains；

For designer based on system noise ρ_l-1Designed symmetric positive definite matrix is at moment t_(l-1)×LValue.

Step is 5.: state observer H based on macroscopic time yardstick_∞F_θMeasurement updaue (Posterior estimator)

Based on step 4. in the systematic parameter of acquisition estimateEstimate with the last step 2. middle system mode obtained Evaluation correctionUtilize the described capacity-SOC-OCV three-dimensional response surface, again update current system open-circuit voltage U_oc；

Based on the above-mentioned current system open-circuit voltage U again updated_oc, electric current u under current sampling point_k=i_L,kAnd voltage Value y_k=U_L,k, carry out current capacities and estimate the correction of the systematic parameter estimated value under macroscopic view sampled point.

Wherein as shown in formula (6), U_ocWith directly determineSize, utilize U_ocCompletion system state is estimated Meter newly ceases renewal, sees formula (27).

Systematic parameter is estimated newly to cease renewal:

Systematic parameter gain matrix:

$K_l^{\mathrm{\θ}}=P_l^{\mathrm{\θ},-}{(I-{\mathrm{\δ}}_{\mathrm{\θ}}{S}_l^{\mathrm{\θ}}{P}_l^{\mathrm{\θ},-}+{\left(C_l^{\mathrm{\θ}}\right)}^T{\left(R_l^{\mathrm{\θ}}\right)}^{-1}{C}_l^{\mathrm{\θ}}{P}_l^{\mathrm{\θ},-})}^{-1}{\left(C_l^{\mathrm{\θ}}\right)}^T{\left(R_l^{\mathrm{\θ}}\right)}^{-1}---\left(28\right)$

Estimate in systematic parameterOn the basis of, utilize systematic parameter to estimate newly to cease renewalWith systematic parameter gain matrixRightIt is modified, obtains system state estimation value correction

Systematic parameter estimated value correction:

Parameter designing matrix update:

Wherein, the System Parameter Design matrix utilizing step 4. to obtain is estimatedThe survey of completion system parameter designing matrix Amount updatesDescribed System Parameter Design matrix updateIt is used for updating the parameter gain matrix of subsequent timeSee formula (26) with (28).

Newly cease for state estimation, i.e. the predictor error of measured value；

The same observation function representing model；

For parameter gain matrix；

It is similarly designer based on measuring noise v_kDesigned symmetric positive definite matrix is at t_kThe value in moment；

After above-mentioned five steps, it is thus achieved that t_kTime inscribe systematic parameter estimated value correctionRepair with system state estimation value JustThen accordingly result will return to step 1. in, and as subsequent time t_k+1At the beginning of lower parameter and state estimation Value.

3, SOC Yu SOH extracts

Based on above-mentioned Multiple Time Scales H_∞Filtering algorithm, obtains real-time battery parameterWith stateBy formula (31) Extract quantity of state z_k, parameter amount C_a,l、R_0,lWith R_p,l。

$\begin{array}{c}{z}_k=\left[\begin{array}{cc}0& 1\end{array}\right]\·{\hat{x}}_k^{+}\\ C_{a,l}=\left[\begin{array}{cccc}0& 0& 0& 1\end{array}\right]\·{\widehat{\mathrm{\θ}}}_l^{+}\\ \left[\begin{array}{c}{R}_{0,l}\\ R_{p,l}\end{array}\right]=\left[\begin{array}{cccc}1& 0& 0& 0\\ 0& 1& 0& 0\end{array}\right]\·{\widehat{\mathrm{\θ}}}_l^{+}\end{array}---\left(31\right)$

C_a,lRepresent t_l×LTime inscribe the battery capacity value of renewal；

R_0,lWith R_p,lRepresent t respectively_l×LTime inscribe the battery ohmic internal resistance of renewal and class value in polarization.

In formula, quantity of state z_kIt is system mode Estimated value correctionIt it is real-time state-of-charge (SOC)；Parameter amount C_a,l、R_0,lAnd R_p,lWith systematic parameter estimated value correctionPhase Close, then can reflect the health status (SOH) of battery the most in real time.

Under a certain fixing service condition (such as temperature constant etc.), battery capacity is the least, and reaction cell is aging the most serious, with Time mean that cell health state (SOH) is the poorest, in this algorithmic procedure, battery capacity precision is higher, can be in this, as master The SOH parameter of measurement wanted；For certain use condition (as temperature constant, SOC fix), the internal resistance of cell is the biggest, Cell degradation is the most serious, again means that battery SOH is the poorest, it is contemplated that internal resistance of cell estimated accuracy is fully checked, because of And only as the auxiliary parameter of measurement of SOH.

Test below by selecting as a example by a certain model nickel-cobalt-manganese ternary battery, so the comparison present invention based on many The estimated value of time scale is relative to estimated value based on single time scale.

Selecting nickel-cobalt-manganese ternary battery is object of study, and its rated capacity is 2Ah, and discharge and recharge blanking voltage is respectively 4.1V、3.0V.Prepare test and include the underlay capacity under three fixed temperature points (10 DEG C, 25 DEG C, 45 DEG C), open-circuit voltage, DST (Dynamic Stress Test, ambulatory stress test operating mode) state of cyclic operation three test, and the basis appearance under room temperature condition Amount is tested with DST state of cyclic operation.Wherein, the test under three fixed temperature points is mainly used in capacity, SOC Yu OCV three's function The debugging of the acquisition of relation, SOC and capacity Combined estimator algorithm routine；Test under room temperature condition is then used for the essence of verification algorithm Degree and stability.

Test based on underlay capacity, obtain the maximum available under different temperatures, as shown in table 1.

Under table 1 different temperatures, this battery cell maximum available

Based on open-circuit voltage (here as a example by temperature) under different capabilities and SOC relation curve, set up battery capacity- The SOC-OCV three-dimensional response surface, as shown in Figure 3.By to capacity, SOC and OCV triadic relation, formula (1), (2) are used to intend Close, thus obtain constant matrices Λ.

$\mathrm{\Λ}=\left[\begin{array}{ccc}22.37& -101.24& 118.36\\ -57.39& 260.98& -296.45\\ 56.30& -258.66& 296.39\\ -23.49& 108.48& -124.48\\ 0.01285& -0.05541& 0.05983\\ 8.470& -37.840& 42.385\\ -0.1370& 0.5997& -0.6541\end{array}\right]---\left(33\right)$

Based on above-mentioned test data and formula (32), by above-mentioned Multiple Time Scales H_∞Filtering algorithm realizes SOC and capacity Combined estimator.Detailed process is:

First, the debugging of Combined estimator algorithm routine is completed.I.e. under three fixed temperature points (10 DEG C, 25 DEG C, 45 DEG C), Based on corresponding DST test data, jointly complete above-mentioned based on double H_∞The SOC of filtering and the tune of capacity Combined estimator algorithm routine Examination.

Then DST test data under room temperature is directly called in the Combined estimator algorithm routine that above-mentioned debugging is good, and will calculate In method, SOC initial value is set to 80% (accurate initial value is 100%), capacity initial value is set to 1.5Ah (accurate initial value is 2.096Ah), obtain room temperature to place an order SOC and capacity estimation result under time scale.

In order to embody the advantage of Multiple Time Scales, take spatial scaling limit value L=1s (single scale) and L=60s (many chis here Degree) two kinds of situations complete electrokinetic cell SOC and capacity estimation, and estimated result is the most as shown in Figure 4, Figure 5.Single scale table in Fig. 5 Show the double H of single time scale_∞The corresponding estimated result of filtering algorithm, curve that is multiple dimensioned and that do not particularly point out all represents chi of many time The double H of degree_∞The corresponding estimated result of filtering algorithm.

Meanwhile, table 2 gives numeral comparing result intuitively.It should be noted that owing to above-mentioned all tests (include not Underlay capacity test, open voltage test and the test of DST state of cyclic operation under synthermal) data sampling time is 1s, works as L During=1s, Multiple Time Scales algorithm deteriorates to single time scale algorithm.

The single time scale of table 2 H double with Multiple Time Scales_∞Filtering estimated result contrast

Note: terminal voltage error is the difference of battery model terminal voltage predictive value and measured value.Measured value is i.e. by test directly Measurement obtains, under the test situation that acquisition precision is high, i.e. it is believed that it is approximately equal to terminal voltage exact value.

SOC error is the difference of above-mentioned filtering algorithm SOC estimation and SOC test reference value.Described SOC tests reference value base Obtaining in following principle: under test conditions, device sensor precision is the highest, thus ensure that current/voltage acquisition precision very High.Under certain experimental condition, charge according to the electric current (generally 0.3C) of standard and battery capacity is full of, now battery SOC is equal to 100%, if needing to start test from 90%SOC, then can bleed off the electric current of 10%SOC according to normalized current, press Method can obtain initial SOC value accurately like this.Carry out correlation test afterwards, during until having tested, according to the electricity of standard Stream is discharged to 0%SOC, SOC value when can have been tested accurately.Known initial with terminate SOC under conditions of, mirror High, the current charge-discharge electrical efficiency in current acquisition precision is it is known that thus use ampere-hour integration method can obtain high-precision SOC with reference to bent Line.

Volume error is the difference of above-mentioned filtering algorithm capacity estimation value and test reference value.Capacity test reference value is Under certain condition, charge according to the electric current of standard after battery capacity is full of, equally battery is discharged to 0% by normalized current Total electricity released in the whole process of SOC.

Draw from above-mentioned analysis, vehicle mounted dynamic battery power system capacity based on data-driven proposed by the invention and SOC Combined estimation method has the advantage that compared with traditional method

(1) expected value that SOC estimates is usedWith the limits of error (as a example by 95% confidence interval, but being not limited to 95%) more Add the possible distribution situation of state-of-charge evaluating electrokinetic cell comprehensively, exactly；

(2) Fig. 4, Fig. 5 show that all energy rapid convergences are to true in the case of SOC with capacity initial value the most inaccurate (20% error) Value, when i.e. achieving battery heap(ed) capacity the unknown, the accurate estimation of SOC, solve tradition SOC algorithm for estimating and hold so that maximum is available Amount be known as premise and cannot Successful utilization to the difficult problem on real vehicle；

(3) battery capacity and internal resistance are to weigh cell health state (SOH) important indicator, thus above-mentioned Combined estimator algorithm Achieve the Combined estimator of SOC Yu SOH to a certain extent；

(4) propose capacity and the SOC algorithm for estimating of Multiple Time Scales, for comparing single time scale algorithm, not only estimate Precision is greatly improved, and decreases algorithm amount of calculation and calculating time significantly.

Claims (9)

1. a Combined estimator electrokinetic cell system state-of-charge and the method for health status, it is characterised in that: described method bag Include:

First, setting up the capacity-SOC-OCV three-dimensional response surface, described OCV is the open-circuit voltage of described system；

Secondly, online data obtains, the voltage and current of Real-time Collection electrokinetic cell system；

Then, Multiple Time Scales filtering algorithm, it is thus achieved that the system under the current macroscopic time scale of described electrokinetic cell system is held Amount advance estimate modification value and the system SOC advance estimate modification value under current microcosmic time scale；

Under each SOC estimates microcosmic sampled point, update described system SOC advance estimate modification value, estimate microcosmic every L described SOC Sampled point is capacity estimation macroscopic view sampled point, updates described power system capacity advance estimate modification under this capacity estimation macroscopic view sampled point Value, the described power system capacity advance estimate modification value after every time updating is as the renewal institute of L time after current capacities estimation macroscopic view sampled point State the parameter used by system SOC advance estimate modification value；

Described L is two or more；

Finally, online SOC Yu SOH extracts, and presently described system SOC utilizing described Multiple Time Scales filtering algorithm to obtain is estimated Correction value and power system capacity advance estimate modification value, estimated driving force battery system state-of-charge and health status.

2. the method for claim 1, it is characterised in that:

Described Multiple Time Scales filtering algorithm includes:

Step is 1.: carries out current SOC and estimates that system SOC under microcosmic sampled point k is estimated, obtains the discreet value of system SOC；

Step is 2.: based on step 1. in the system SOC discreet value of acquisition and the system of last capacity estimation macroscopic view sampled point hold Amount advance estimate modification value, utilizes the described capacity-SOC-OCV three-dimensional response surface, updates current system open-circuit voltage and obtains the first open circuit Voltage；

Then, based on above-mentioned first open-circuit voltage, carry out the correction of described system SOC discreet value, obtain described system SOC and estimate Correction value；

Step is 3.: k+1 estimates microcosmic sampled point as new SOC, it is judged that whether k-1 can be divided exactly by L, if it can, then walk The most 4.；Otherwise return step 1.；

Step is 4.: the power system capacity carried out under capacity estimation macroscopic view sampled point l+1 is estimated, and obtains power system capacity discreet value；

Step is 5.: based on step 4. in the power system capacity discreet value of acquisition and the last step 2. in system SOC that obtains Advance estimate modification value, utilizes the described capacity-SOC-OCV three-dimensional response surface, again updates current system open-circuit voltage and obtains second and open Road voltage；

Based on described second open-circuit voltage, carry out the correction of power system capacity discreet value under capacity estimation macroscopic view sampled point l+1, To power system capacity advance estimate modification value；And return step 1..

3. method as claimed in claim 2, it is characterised in that: step 1. in utilize SOC estimate under microcosmic sampled point k-1 be System current value and system SOC advance estimate modification value, and the power system capacity advance estimate modification value of the last sampled point l carries out system SOC estimates.

4. method as claimed in claim 2, it is characterised in that: step 2. in carry out the correction of described system SOC discreet value, base In step 1. in the system SOC discreet value of acquisition, described first open-circuit voltage, SOC estimate the current value under microcosmic sampled point k And magnitude of voltage.

5. method as claimed in claim 2, it is characterised in that: step 5. in carry out the correction of described power system capacity discreet value, The described system under microcosmic sampled point is estimated based on the second open-circuit voltage, capacity estimation macroscopic view sampled point l+1 or the last SOC Current value and magnitude of voltage.

6. method as claimed in claim 1 or 2, it is characterised in that: utilize power system capacity advance estimate modification value to assess described SOH.

7. method as claimed in claim 1 or 2, it is characterised in that: described SOC estimates sample frequency and the institute of microcosmic sampled point The sample frequency stating electric current and/or voltage is identical.

8. method as claimed in claim 1 or 2, it is characterised in that: set up the side of described capacity-SOC-OCV three-dimensional response surface Method is: be fitted SOC with the OCV relation under different capabilities, obtains the built-up pattern coefficient under each described different capabilities, adopts With quadratic function, the relation of each built-up pattern coefficient with capacity is fitted, completes building of capacity-SOC-OCV three-dimensional response surface Vertical.

9. method as claimed in claim 1 or 2, it is characterised in that: by battery management system Real-time Collection electrokinetic cell system Voltage, electric current and/or the temperature of system.